Modelpublic

FunAudioLLM/SenseVoiceSmall

sourceHugging Faceotherupdated 3mo agoView on Hugging Face

483likes24kdownloads

demo.py28 linesDownload Raw Back to root

1from funasr import AutoModel2from funasr.utils.postprocess_utils import rich_transcription_postprocess3 4model_dir = "FunAudioLLM/SenseVoiceSmall"5 6 7model = AutoModel(8    model=model_dir,9    vad_model="fsmn-vad",10    vad_kwargs={"max_single_segment_time": 30000},11    device="cuda:0",12    hub="hf",13)14 15# en16res = model.generate(17    input=f"{model.model_path}/example/en.mp3",18    cache={},19    language="auto",  # "zn", "en", "yue", "ja", "ko", "nospeech"20    use_itn=True,21    batch_size_s=60,22    merge_vad=True,  #23    merge_length_s=15,24)25text = rich_transcription_postprocess(res[0]["text"])26print(text)27 28